def _start_restricted_worker(self, num_cores=0):
command = RayServiceFuncGenerator._get_raylet_command(
redis_address=self.redis_address,
ray_exec="ray ",
password=self.redis_password,
ray_node_cpu_cores=num_cores,
object_store_memory=self.object_store_memory,
extra_params=self.extra_params)
print("Executing command: {}".format(command))
process_info = session_execute(command=command, fail_fast=True)
ProcessMonitor.register_shutdown_hook(pgid=process_info.pgid)
def _start_cluster(self):
ray_rdd = self.sc.range(0,
self.num_ray_nodes,
numSlices=self.num_ray_nodes)
from zoo import ZooContext
if ZooContext.barrier_mode:
print("Launching Ray on cluster with Spark barrier mode")
# The first ip would be used to launch ray master.
process_infos = ray_rdd.barrier().mapPartitions(
self.ray_service.gen_ray_start(self.cluster_ips[0])).collect()
else:
print("Launching Ray on cluster without Spark barrier mode")
master_process_infos = ray_rdd.mapPartitionsWithIndex(
self.ray_service.gen_ray_master_start()).collect()
master_process_infos = [
process for process in master_process_infos if process
]
assert len(master_process_infos) == 1
master_process_info = master_process_infos[0]
redis_address = master_process_info.master_addr
raylet_process_infos = ray_rdd.mapPartitions(
self.ray_service.gen_raylet_start(redis_address)).collect()
raylet_process_infos = [
process for process in raylet_process_infos if process
]
assert len(raylet_process_infos) == self.num_ray_nodes - 1
process_infos = master_process_infos + raylet_process_infos
self.ray_processesMonitor = ProcessMonitor(process_infos,
self.sc,
ray_rdd,
self,
verbose=self.verbose)
return self.ray_processesMonitor.master.master_addr
def _start_restricted_worker(self, num_cores, node_ip_address):
extra_param = {"node-ip-address": node_ip_address}
if self.extra_params is not None:
extra_param.update(self.extra_params)
command = RayServiceFuncGenerator._get_raylet_command(
redis_address=self.redis_address,
ray_exec="ray",
password=self.redis_password,
ray_node_cpu_cores=num_cores,
object_store_memory=self.object_store_memory,
extra_params=extra_param)
modified_env = self.ray_service._prepare_env()
print("Executing command: {}".format(command))
process_info = session_execute(command=command, env=modified_env,
tag="raylet", fail_fast=True)
ProcessMonitor.register_shutdown_hook(pgid=process_info.pgid)
def _start_cluster(self):
print("Start to launch ray on cluster")
ray_rdd = self.sc.range(0, self.num_ray_nodes,
numSlices=self.num_ray_nodes)
process_infos = ray_rdd.barrier().mapPartitions(
self.ray_service.gen_ray_start()).collect()
self.ray_processesMonitor = ProcessMonitor(process_infos, self.sc, ray_rdd, self,
verbose=self.verbose)
self.redis_address = self.ray_processesMonitor.master.master_addr
return self
def _start_cluster(self):
print("Start to launch ray on cluster")
ray_rdd = self.sc.range(0, self.num_ray_nodes,
numSlices=self.num_ray_nodes)
# The first ip would be used to launch ray master.
process_infos = ray_rdd.barrier().mapPartitions(
self.ray_service.gen_ray_start(self.cluster_ips[0])).collect()
self.ray_processesMonitor = ProcessMonitor(process_infos, self.sc, ray_rdd, self,
verbose=self.verbose)
return self.ray_processesMonitor.master.master_addr